Assessing the accuracy of soil organic carbon estimation in agricultural fields using satellite and UAV images under semiarid conditions

• Soil organic carbon (SOC) was predicted for completely crop-uncovered and partially crop-covered soils using different remote sensing platforms. • A single prediction model may be better for global application, but the accuracy improves when different types of soil coverage are considered. • Predictors, such as terrain attributes, electrical conductivity (EC) and crop development are considered relevant. Soil organic carbon (SOC) estimation and mapping are the primary pillars of sustainable soil management. Traditionally, field sampling for SOC determination requires a significant amount of time and human resources. Therefore, analysing the accuracy of remote sensing techniques at different resolutions for SOC prediction is crucial. This study proposes a multi-source data approach that incorporates remote sensing data from both medium (satellites) and very-high (uncrewed aerial vehicles) resolution sources for bare soil, vegetation indices of crop canopy, soil measurements and terrain characteristics at a plot level in a semiarid environment. Furthermore, the study aims to evaluate the accuracy of these predictions in two distinct agricultural contexts: plots with permanent tree crops, where completely bare soil is not visible throughout the growing season, and plots with annual crops, i.e., those that complete their growth and development within a single growing season, where the soil is exposed during some periods of the year. Results showed that better adjustments are obtained when plots with permanent tree crops and plots with annual crops are considered separately: R 2 adj = 0.325 for both types of bare soil vs. R 2 adj = 0.424 and 0.561 for permanent tree crops and annual crops, respectively. Additionally, using normalised vegetation index (NDVI) values from the crop at different stages of the growing cycle enhances the prediction’s accuracy.